Processor and motherboard
Multithreading is supported by all contemporary engineering analysis software solvers; however, because of the peculiarities of the problems to be solved, it is preferable to scale them by increasing the clock frequency and processor core performance rather than by increasing the number of cores or threads.
The LIRA-FEM solver is not an exception. It also responds better to an increase in the performance of cores rather than to an increase in number of cores. High utilisation of all cores/threads is achieved only at some analysis stages. For example, when decomposing the stiffness matrix, calculating the DCF/DCL, and analysing reinforcement.
Therefore, LIRA-FEM is better suited for desktop platforms and desktop series processors. On Xeon solvers, it is not logical to use workstations and servers for LIRA-FEM analysis. For other problems, these solvers are optimised. Compared to desktop series solvers, these solvers are slower.
Budget computer for analysis of small problems
Any contemporary processor with a comparatively high frequency can be suggested for comfortable work when performing analysis of simple problems.
For AMD platform these are 4- and 6-core RYZEN 3 or RYZEN 5 processors. Almost any motherboard with AM4/AM5 socket is suitable for these processors.
For Intel platform it can be 4-core Core i3 12000/13000/14000 processors. These processors have a built-in graphics core (except for the F series), but its performance when working with 3D graphics will not be enough. Almost any motherboard with LGA 1700 socket will be suitable for such a processor.
Powerful computer to analyse large problems
For large problems, the processors of the higher families are preferable. These processors have high operating frequencies in addition to the increased number of cores. To obtain the best performance, you should focus on them. Also, the use of such processors can increase the overall performance of the system under load.
For the AMD platform, it will be senior processors of the RYZEN 7000/8000/9000 family, for example, the RYZEN 7 7700X and a corresponding motherboard with an AM5 socket. Use RYZEN 9 processors, but remember that analysis speed does not increase linearly with core count, and if there are more than 16 cores, speed may not increase.
For the Intel platform, these are Core i7/i9 processors, for example, the Core i7-12700/13700/14700K (the features of these processors are described below). A motherboard with an LGA 1700 socket and a potent enough CPU power supply system are necessary for these processors.
Be careful: modern processors of previous series with 8 or more cores are characterised by high power consumption and heat generation. For such processors, you have to choose motherboards with powerful 9 or more channel CPU power supply systems. It is possible to use boards of game series, as in them manufacturers pay much attention to the CPU power supply system.
The cooling system for processors of the previous series should also be selected from the most productive, with a dissipated power of 180 W or more. Since Ryzen 9 and Core i9 processors can require up to 240W at full load and an air cooling system might not be able to handle so much heat, it might be justified to use liquid cooling systems.
Hybrid processors Intel 12/13/14 series, Windows 10 and LIRA-FEM
Intel CPUs with hybrid architectures from the 12, 13, 14, and Ultra series are now available on the market. Its meaning is that the processor has high-performance (large) cores and also "energy-efficient" (small) cores that perform significantly less than the large cores. This is done to limit the overall power consumption of the processor while the number of cores is increased. Every Intel i7/i9, Ultra7/Ultra9 CPU in this series has eight large cores, while the number of small cores varies from four to sixteen. The small cores are about 5 times weaker than the large cores.
When using Windows 10, the operating system scheduler allocates resources of large cores only to the running application. If the application is inactive (minimise the application window and do something else), the application will be switched to small cores, and it will cause the analysis speed to drop several times. For example, our test problem on i7-12700K (8 large, 4 small cores) was analysed for 10 minutes in active mode and for 85 minutes in inactive mode
So, the recommendation for the Intel platform looks like this:
- There is no point in buying an i9 12-14 series processor; there are only 8 large cores there anyway. It would make sense to go with i7 12700/13700/14700K.
- It is possible not to use expensive DDR5 memory and choose a board with DDR4 support; the analysis speed will practically remain the same.
- Use Windows 11 (it fixes this this issue) or do nothing on the computer during the analysis or disable small cores in BIOS/UEFI settings.
RAM
The issues and applications you use will determine how much RAM you need. A big memory might provide a speed advantage for certain analysis steps, including stiffness matrix decomposition or some iterative calculations.
The min recommended RAM for the LIRA-FEM application and its LIRA-CAD module is 16GB. RAM between 32 and 64 GB is recommended if you are going to analyse large models, use 3D modelling software, or work with more complex CAD/CAM/CAE systems.
Disc system
The speed of work in the LIRA-FEM program also depends on the disc system. When analysing highly complex models, the disc is heavily accessed, particularly when iterative analysis techniques are employed. The application writes a lot of data to its working directories, DATA (input data files) and WORK (analysis results, temporary files).
The industry standard at the moment is solid-state drives (SSDs). The analysis speed can be considerably decreased by using mechanical hard drives for the DATA and WORK working directories.
Care should be taken when selecting an SSD. For example, many M-2 form factor drives are likely to overheat if placed in locations with poor air circulation. Budget series drives, particularly QLC, can reduce write speeds to zero under heavy load, which can cause issues when analysing the large models. The professional series of SSDs is the recommended option.
Remember that over time, the WORK directory accumulates quite a lot of data. This is because engineers usually create several versions of model files during their work and save the analysis results for these versions. It is advisable to monitor this and refrain from overloading the SSD with unnecessary data. Because of the nature of the technology, filling SSDs to more than 50–60% will cause them to wear down much more quickly and may also result in a considerable decrease in write speed.
Video system
A GeForce GTX 1650 or GeForce GTX 1660 Ti video card is adequate for the LIRA-FEM application and its LIRA-CAD module to function comfortably. More powerful so-called "gaming" graphics cards, like the GeForce RTX 4060, are highly encouraged.
The relatively inexpensive Quadro T1000 professional graphics card also performed well. Although it is three times slower than the 4060, it has half the power consumption. Intel processors have an integrated Intel HD video core. The performance of this core is not sufficient for comfortable work. Integrated video APU AMD is completely inappropriate for our purposes. If you choose a computer with an AMD processor, a discrete graphics card is required.
Casing, power supply
A high-quality casing and power supply are recommended. Poor ventilation can cause overheating, and a poor-quality power supply unit frequently malfunctions, particularly when used over extended periods of time. The dust filter in the casing is, in our opinion, a big benefit.
The power supply should be selected with some reserve if it will be used for extended periods of time. A 500–700 W power supply is usually adequate for a work station. If you want to use a powerful video card with a consumption of more than 150 W, you'll need a higher power supply.
Laptop for analysis
Laptops are obviously less productive than comparable desktop systems, but if there is a need to use a laptop for analysis, the optimal choice is the AMD RYZEN 5/7 7000 series processor. These processors have good performance, but their built-in video core is completely unsuitable for LIRA-FEM application and its LIRA-CAD module. Therefore, you should only consider laptops that have a discrete graphics card such as the GeForce GTX 1660 Ti Mobile, GeForce RTX 3050 Ti Mobile, GeForce RTX 3070 Ti Mobile or similar. Cards labelled Ti are generally preferred as they offer better performance.
A discrete graphics card is required for comfortable work if you select a laptop with an Intel processor. Although leisurely work with small models is also possible on the video adapter built into the processor.
Since laptops rarely come with a lot of RAM, you should think about upgrading the memory when you purchase a laptop. There is only one memory module slot available on many contemporary laptops.
Memory in modern slim series laptops may be soldered to the motherboard and cannot be expanded. When making a purchase, this should be considered.
We do not recommend laptops with high-resolution screens (4K, HiDPI) when working with CAD systems. When working with images or videos, these screens might be helpful, but they won't display CAD system interfaces — like the LIRA-FEM program — at their best. The one exception is if you plan to use a laptop equipped with an external 4K monitor as your primary monitor. In this case, the built-in monitor of the laptop can have any resolution and can be used only as an auxiliary monitor.
HP, Dell, Lenovo workstations
Computer manufacturers provide tailored workstation configurations or off-the-shelf options. These workstations are not optimised for finite element analysis (FEA), which is one of their peculiarities.
First of all, workstations with Core i7/i9 processors rarely have adequate RAM. The installed RAM is usually 16 GB, which might not be sufficient for complex computations. As previously stated, workstations equipped with Xeon processors have 32 GB of installed memory; but, due to the slower performance of FEA, this memory is not being used to its full potential.
Secondly, workstations with Core i7/i9 CPUs can have no discrete visual adapters at all, but the Intel video has poor performance.
Thirdly, when used extensively as a structural engineer's workstation, these laptops may not have adequate space because they typically have 256–512 GB SSDs and no larger HDD. You can free up space on the SSD by storing problem archives with analysis results on a 2TB HDD.
Thus, there are almost no ready -made ideal workstations for FEA. The solution is to purchase customised configurations with roughly the following components:
- Dell OptiPlex / Lenovo ThinkStation / HP Z with Core i9 or i7 processor
- 2 TB SSD М.2. 4 TB HDD, if archives are not supposed to be stored on the network.
- NVidia GeForce RTX 4060 8 Gb
- 32 GB RAM
If you find a mistake and want to inform us about it, select the mistake, then hold down the CTRL key and click ENTER.
Comments 1